HotSoS 2024 Summary Report
The National Security Agency (NSA) virtually hosted the 11th Annual Symposium on Hot Topics in the Science of Security (HotSoS), from 2-4 April 2024. The General Chair was Adam Tagert, (NSA), and Program Co-Chairs were Alexander Velazquez, Naval Research Lab (NRL), and Adam Petz (University of Kansas). HotSoS brings together researchers from diverse disciplines to promote the advancement of work related to the Science of Security initiative (SoS), and features a mix of invited keynotes, Works-in-Progress (WiP) discussions, presentations of already published work, posters, and panel discussions.
Almost 300 individuals attended HotSoS 2024, and participants were a mix of government, academia, and industry from the US, Canada, the UK, Austria, Germany, and India. In addition to three keynote presentations, HotSoS 2024 included a presentation by the winner of NSA's 11th Annual Best Scientific Cybersecurity Paper Competition, eight published papers, six WiP manuscripts, and 19 posters or demos. The papers and posters represented the work of 59 authors from 41 universities and institutions. In keeping with the goal of collaborative community engagement, HotSoS 2024 again featured WiPs, which provide an opportunity for authors to get early feedback on a research direction, technology, or ideas before a paper has been fully evaluated, as well as to discuss systems in an early, pre-prototyping phase.
Following introductory remarks, Dr. Rita Bush, Chief of NSA’s Laboratory for Advanced Cybersecurity Research (LACR), welcomed the attendees and noted the recent award of three Virtual Institutes (VIs) in the areas of Artificial Intelligence (AI) and Cybersecurity, Defensive Mechanisms, and Trusted Systems. The VIs cover 11 projects at seven different universities.
More information on the VIs can be found here.
Keynote Presentations
The first keynote presentation, “Language Models for Formal Proof,” was given by Talia Ringer of the University of Illinois at Urbana-Champaign. They first defined formal proof as proofs that are developed using tools known as proof assistants and then described how the tools work. Proof automation makes it easier to develop and maintain verified systems using proof assistants, but despite advances in these tools over the past decades, there is still a lot of work to be done. They noted that traditional automation tools are predictable, dependable, and understandable, but are limited in scope and require expertise to extend. On the other hand, they noted, building automation using language models demonstrate opposite characteristics—unpredictable, not dependable, and not understandable but aren’t limited in scope and are easily extendable. The goal would be to get the best of both worlds, and their presentation described how the different methods can come together to make writing machine-checkable proofs of program correctness easier, and what this means for the future of machine-checkable proof.
Stephen Smalley (NSA) gave a keynote presentation entitled “Retrospective: 30 years of Cybersecurity R&D,” in which he spoke about NSA’s LACR, noting its long history of open-source contribution and collaboration, and the impacts its work has had on both open-source and proprietary systems. He cited two recurring themes for LACR research: microkernels for security and security for microkernels; and flexible security/Mandatory Access Control. He discussed technological advancements over the past 30 years and what those advancements have meant for LACR research focus areas.
The final keynote was given by Aaron Weissenfluh of Tenfold Security and was entitled “Defeating Ransomware: How a Small Midwestern University Defeated Ransomware.” This presentation detailed a ransomware incident and response that was experienced by a small university over a 10-day period just before the start of the fall term. He noted that the university had a very small staff with no cybersecurity personnel, but they called in his company and did what was recommended. Mr. Weissenfluh described activities that were undertaken so that systems were restored before the students arrived on campus without a ransom being paid. He also listed the procedures that have been put in place by the university since the incident, identifying the top lessons-learned as awareness, communication, and preparation and prevention.
Best Scientific Cybersecurity Paper Competition Invited Talk
The winning paper in the SoS 11th Annual Best Scientific Cybersecurity Paper Competition was “Uninvited Guests: Analyzing the Identity and Behavior of Certificate Transparency Bots” by Nick Nikiforakis, Brian Kondracki, and Johhny So of Stony Brook University. Professor Nikiforakis gave a presentation entitled “Bots, Crawlers, and Spiders: Understanding How Automated Web Clients Find and Attack their Victims,” which provided a background for their study of malicious bots and the reasoning behind their decision to write the paper as well as a discussion of the paper itself. The research question was whether they could curate a bot-only dataset that doesn’t depend on their manual analysis prowess. The paper examined a study of automated attacks on new webservers and explored how a web browser can trust an organization’s publicly issued cryptographic credentials. The researchers studied autonomous systems which probe newly instantiated encrypted websites. They identified 105 malicious security bots attempting to perform nefarious actions such as data exfiltration, reconnaissance, and vulnerability exploitation. They also identified security systems examining sites to identify new phishing attacks. These profiles provide new insights into these autonomous actions happening on the Internet. This data can be used by both system administrators and developers to protect systems from compromise. The research team collected this data by creating the Certificate Transparency Honeypot (CTPOT), a system that obtains new certificates and monitors web bots for potential targets. CTPOT allows researchers to trick web bots, isolate them, and identify if they are malicious.
Works-Already-Published Sessions
The Works-Already-Published sessions featured eight published papers.
1. Adversarial Data-Augmented Resilient Intrusion Detection System for Unmanned Aerial Vehicles
Muneeba Asif, Florida International University (FIU)
This work proposes a novel Intrusion Detection System (IDS) for UAVs that enhances resilience against adversarial Machine Learning (ML)-based attacks by using Generative Adversarial Networks (GAN). The authors also studied several evasion-based adversarial attacks and used them to compare the performance of the proposed IDS with existing ones. The evaluation results demonstrate that the proposed IDS is significantly robust against adversarial ML-based attacks compared to the state-of-the-art IDSs while maintaining a low false positive rate.
2. SyzDescribe: Principled, Automated, Static Generation of Syscall Descriptions for Kernel Drivers
Yu Hao, University of California, Riverside
Fuzz testing operating system kernels has been effective overall in recent years, but syscall descriptions are largely written manually. This paper presents a principled solution for generating syscall descriptions for Linux kernel drivers. The authors summarize and model the key invariants or programming conventions, extracted from the “contract” between the core kernel and drivers, allowing them to understand programmatically how a kernel driver is initialized and how its associated interfaces are constructed. They developed a tool called SyzDescribe and showed that the syscall descriptions produced by SyzDescribe are competitive to manually-curated ones.
3. Formal Specification and Verification of Architecturally-defined Attestation Mechanisms in Arm CCA and Intel TDX
Muhammad Usama Sardar, TU Dresden
This work presents a holistic verification approach enabling comprehensive and rigorous security analysis of architecturally-defined attestation mechanisms in Confidential Computing. The authors analyzed two prominent next-generation hardware-based Trusted Execution Environments (TEEs), specifically Arm Confidential Compute Architecture (CCA) and Intel Trust Domain Extensions (TDX), and provided a comprehensive specification of all phases of the attestation mechanism, namely provisioning, initialization, and attestation protocol. They demonstrate that including the initialization phase in the formal model leads to a violation of integrity, freshness, and secrecy properties for Intel’s claimed Trusted Computing Base (TCB), which could not be captured by considering the attestation protocol alone in the related work.
4. Security Misconfigurations in Open Source Kubernetes Manifests: An Empirical Study
Akond Rahman, Auburn University
This paper presents the results of an empirical study with 2,039 Kubernetes manifests mined from 92 open-source software repositories to systematically characterize security misconfigurations in Kubernetes manifests. The researchers also constructed a static analysis tool called Security Linter for Kubernetes Manifests (SLI-KUBE) to quantify the frequency of the identified security misconfigurations. They identified 11 categories of security misconfigurations, and observed that the identified security misconfigurations affected entities that perform mesh-related load balancing, as well as provision pods and stateful applications. The research showed the need for security-focused code reviews and the application of static analysis when Kubernetes manifests are developed.
5. An Attack Volume Metric
Massimiliano Albanese, George Mason University (GMU)
Various metrics have been developed to quantify the extent of a system's attack surface, but most approaches have failed to consider the complex interdependencies that exist between the many components of a distributed system, its vulnerabilities, and its configuration parameters. This work builds upon previous research on vulnerability metrics and on graphical models to capture such interdependencies, and proposes a novel approach to evaluate the potential risk associated with exposed vulnerabilities by studying how the effect of each vulnerability exploit propagates through chains of dependencies. The analysis goes beyond the scope of traditional attack surface metrics, and considers the depth and implications of potential attacks, leading to the definition of a new family of metrics, which the authors refer to as attack volume metrics. The paper presents experimental results illustrating how the proposed metric scales for graphs of realistic sizes, illustrating its application to real‐world testbeds.
6. A Tale of Two Industroyers: It was the Season of Darkness
Luis Salazar, University of California, Santa Cruz
The researchers studied two pieces of malware that attempted to create blackouts in Ukraine. They designed and developed a new sandbox that emulates different networks, devices, and other characteristics in order to execute malware targeting substation equipment and understand in detail the specific sequence of actions the attackers could perform on substation equipment. They also studied the effects that similar malware could have in the future. Their findings include new malware behavior not previously documented (such as the detailed algorithm for the MMS protocol payload) and an illustration of how attacking different targets will produce different effects.
7. Building and Testing a Network of Social Trust in an Underground Forum: Robust Connections and Overlapping Criminal Domains
Dalyapraz Manatova, Indiana University
The anonymity of underground marketplaces and incentives to avoid penalties for criminal activity create significant challenges in studying trust in these ecosystems. To contribute to understanding online crime, the authors offer an empirical analysis of an underground forum by examining interactions in the social network as a whole and those components of the network that support three major types of crime: traditional crimes that occur away from keyboards, transitional crimes that have both offline and online instantiations, and entirely online new crimes. The results suggest that although communities follow the small world effect, identifying and removing highly connected moderators or prolific contributors will not harm any of these three communities or the network, unless a significant portion of the network is removed. By further observing the structural patterns, the researchers find that transitional crime actors tend to cluster more compared to the other two crimes while having the highest density.
8. MAYALOK: A Cyber-Deception Hardware Using Runtime Instruction Infusion
Preet Derasari, George Washington University
Cyber-deception is an increasingly adopted defense strategy against malware attacks given its ability to continually engage with adversaries and deploy counter-measures proactively by manipulating the malware program execution flow to non-useful states for the attacker. This paper introduces Mayalok, a hardware-based cyber-deception framework to combat malware through runtime instruction infusion. Mayalok employs hardware deception primitives to transparently insert or skip malware program instructions during runtime and deliver the attackers a deceptive view of the system state.
Works-in-Progress Sessions
The Works-in-Progress sessions featured six papers.
1. Large Language Models for Static Analysis
Yu Hao (University of California, Riverside)
2. Studying Cyber Behavior with Honeypots
Jingyang Zou, Zhaoxi Sun, and Chunyen Ku (Johns Hopkins University)
3. Deep Graph Neural Networks for Malware Detection using Ghidra P-Code
Rinaldo Iorizzo (Rochester Institute of Technology)
4. Understanding Vulnerability Discovery in Expert and Novice Binary Analysts’ Behavior
K. Raghav Bhat (Arizona State University)
5. Rigorously Tested Cyber Deception Technologies
Jason Landsborough (Naval Postgraduate School)
6. Autonomous Cyber Defense using Evolving behavior trees
Hunter Bergstrom (Vanderbilt)
Panel Discussions
Best Practices for Doing/Reviewing Research Papers in a Program Committee
Yan Shoshitaishvili (Arizona State University), Anna Marie Ortloff (University of Bonn), Arun Vishwanath (Avant Research Group, LLC), Jeffrey Carver (University of Alabama)
The panelists provided their insights both from the perspective of reviewers and as paper authors. They noted that the core of the review process is subjective: the definition of novelty, rigor, and scientific methodologies are subjective. They agreed that the process is universally hated by both submitters and reviewers, but there is no better process. In reviewing papers, they noted that it’s much easier to find reasons for rejecting a paper than it is for accepting it. They suggested that authors think of their paper across two axes: papers that have to solve a problem and papers that identify problems and find a solution; reviewers are always looking for a solution that is understandable as well as replicable. They said that reviewers needed to keep their egos out of the review, to review the paper the author wrote rather than the one the reviewer thought they should have written, and to consider how they would receive the review. They also recommended active mentorship with experienced reviewers, giving new reviewers some insight.
Future of Foundational Research
Neal Ziring (NSA), Evan Austin (NRL), Morgan Stern (NSA), and Cathie Cope (NSA)
The panelists addressed the effects of Large Language Models (LLMs) on a societal level, given that society is still grappling with LLMs, and the fact that societal changes will make it so LLMs will be adopted more. They noted that LLMs are a new attack vector that threat actors will target. The panelists also discussed the risks associated with AI/ML as it becomes more integrated into workflows and cybersecurity. Since many security vulnerabilities are due to the complexity of systems, one panelist posited that a science of secure composition is needed. Another cited the need for more research in formal methods.
Gaps and research challenges for Zero-Trust
Karen Uttecht (Massachusetts Institute of Technology), Tim Morrow (Software Engineering Institute/CERT), Selcuk Uluagac (FIU), Shelly Kelly (NSA)
The panelists identified zero-trust as a strategy and said that it would be the cybersecurity framework for the next 20-30 years. They cited AI/ML as key to the future of zero-trust, noting that it would help with finding gaps and reducing noise in the system. Other research areas included how to measure security, what policies are needed, continuous authority to operate, automation, and adapting zero-trust concepts to newer domains.
Announcement of Winning Poster / Closing Remarks
The HotSoS Best Paper Award was “Utilizing LLMs to Translate RFC Protocol Specification to CPSA Definitions.” It was presented by Martin Duclos, Mississippi State University (MSU), and its co-authors were Ivan Fernandez (MSU), Kaneesha Moore (MSU), Sudip Mittal (MSU), and Edward Zieglar (NSA).
HotSoS 2025 will be held in April 2025 and will again be virtual.
The agenda and selected presentations are available here.
Submitted by Cyber Pack Ventures, Inc.
